BUG: Model configuration error dialog displays long messages outside the container

[jmbillard]

Summary

Long error messages in the model configuration overflow outside the dialog/container, causing part of the text to be rendered off-screen and become unreadable.

Environment

• Windshift self-hosted
• Docker deployment

Steps to Reproduce

1. Open the model configuration page.
2. Configure a model with invalid or incomplete settings.
3. Trigger a validation or save error that produces a long error message.
4. Observe the displayed error message.

Actual Behavior

Long error messages extend beyond the right edge of the dialog/container.

Part of the message is rendered outside the visible area, making it impossible to read the full error.

Expected Behavior

Error messages should remain within the bounds of the dialog/container.

Long messages should either:

• Wrap onto multiple lines.
• Be displayed in a scrollable container.
• Resize the container appropriately so the full message remains visible.

Additional Information

The issue appears to be related to the layout of the error message rather than the error itself.

Instead of wrapping or remaining within the available space, the text overflows horizontally outside the visible UI.

Possible Cause

Possible issues include:

• Missing text wrapping (word-break or overflow-wrap).
• Horizontal overflow is not handled by the container.
• The error component does not constrain long strings to the available width.

Request

Please ensure that long error messages remain fully visible within the model configuration UI and do not overflow outside the visible container.

Thanks!

[stefan-ernst]

Thanks, will parse out the raw message here

One more thing: Is that model working in general? The Test command should not reach max tokens

[jmbillard]

@stefan-ernst
I tested gpt-4o-mini successfully, and the connection test completed without issues.

However, both the new 5.6-luna model and 5.3-chat return the same max tokens error:

Could not finish the message because max_tokens or model output limit was reached. Please try again with higher max_tokens.

Since the test command should not generate a large response, it seems like the issue may be related to how the request is being sent or how the output limit is being handled for these models.

The error is returned as a LiteLLM BadRequestError (status 400) from the LLM API.

[stefan-ernst]

Thanks! I will have a look - Luna 5.6 via Openrouter seems to work, I will try to set up LiteLLM when I have time

[stefan-ernst]

I was able to get it working via LiteLLM and Openrouter:

one question remaining is whether OpenAI API behaves the same way. I will check this too

[stefan-ernst]

Found the issue:

Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead.

its an API contract issue for OpenAI. Will fix in 0.8.5